Authors: Stella Bounareli, Christos Tzelepis, Vasileios Argyriou, Ioannis Patras, Georgios Tzimiropoulos

Published on: February 05, 2024

Impact Score: 8.22

Arxiv code: Arxiv:2402.03553

Summary

• What is new: Uses pre-trained GANs and a 3D shape model for higher quality facial reenactment.
• Why this is important: Previous methods degraded image quality due to complex embedding networks for disentangling identity and head pose/expression.
• What the research proposes: A framework utilizing pre-trained GANs and a 3D shape model to identify and control latent space directions for head pose and expression, supporting high-quality, one-shot, cross-person facial reenactment.
• Results: Produces higher quality reenacted faces compared to state-of-the-art methods on VoxCeleb1 2 benchmarks.

Technical Details

Technological frameworks used: Pre-trained GANs, 3D shape model

Models used: Latent space direction discovery

Data used: VoxCeleb1 2

Potential Impact

Entertainment industry, digital content creation platforms, security and surveillance sectors.

Want to implement this idea in a business?

We have generated a startup concept here: FaceUnity.